Data Types

ndonnx exposes the following data types. There is also unstable support for defining your own structured data types. “Primitive” data types have a directly corresponding data type in the ONNX standard.

Name	Description	Array API	ONNX primitive
float32	32-bit floating point number	Yes	Yes
float64	64-bit floating point number	Yes	Yes
int8	8-bit signed integer	Yes	Yes
int16	16-bit signed integer	Yes	Yes
int32	32-bit signed integer	Yes	Yes
int64	64-bit signed integer	Yes	Yes
uint8	8-bit unsigned integer	Yes	Yes
uint16	16-bit unsigned integer	Yes	Yes
uint32	32-bit unsigned integer	Yes	Yes
uint64	64-bit unsigned integer	Yes	Yes
bool	Boolean	Yes	Yes
utf8	UTF-8 string	No	Yes
nfloat32	Nullable 32-bit floating point number	No	No
nfloat64	Nullable 64-bit floating point number	No	No
nint8	Nullable 8-bit signed integer	No	No
nint16	Nullable 16-bit signed integer	No	No
nint32	Nullable 32-bit signed integer	No	No
nint64	Nullable 64-bit signed integer	No	No
nuint8	Nullable 8-bit unsigned integer	No	No
nuint16	Nullable 16-bit unsigned integer	No	No
nuint32	Nullable 32-bit unsigned integer	No	No
nuint64	Nullable 64-bit unsigned integer	No	No
nbool	Nullable boolean	No	No
nutf8	Nullable UTF-8 string	No	No
datetime64	Datetime data type that follows NumPy’s datetime64 semantics	No	No
timedelta64	Timedelta data type that follows NumPy’s timedelta64 semantics	No	No